5. AF_XDP Poll Mode Driver
AF_XDP is an address family that is optimized for high performance packet processing. AF_XDP sockets enable the possibility for an XDP program to redirect packets to a memory buffer in userspace.
Further information about AF_XDP can be found in the AF_XDP kernel documentation.
This Linux-specific PMD creates the AF_XDP socket and binds it to a specific netdev queue. The DPDK application can then send and receive raw packets through the socket which bypass the kernel network stack.
5.1. Prerequisites
- A Linux Kernel (version >= v4.18) with the XDP sockets configuration option enabled (CONFIG_XDP_SOCKETS=y).
- Both libxdp (>= v1.2.2) and libbpf (any version) libraries installed, or alternatively just the libbpf library <= v0.6.0.
- The pkg-config package should be installed on the system as it is used to discover the libbpf and libxdp libraries and determine their versions are sufficient.
- If using libxdp, it requires an environment variable called LIBXDP_OBJECT_PATH to be set to the location of where libxdp placed its bpf object files. This is usually in /usr/local/lib/bpf or /usr/local/lib64/bpf.
- A Kernel bound interface to attach to.
- The need_wakeup feature requires kernel version >= v5.4.
- The PMD zero copy feature requires kernel version >= v5.4.
- The shared UMEM feature requires kernel version >= v5.10 and libbpf version v0.2.0 or later. The LINUX_VERSION_CODE defined in the version.h kernel header is used to determine the kernel version at compile time.
- A kernel with version 5.4 or later is required for 32-bit OS.
- The busy polling feature requires kernel version >= v5.11.
5.2. Options
5.2.1. iface
The iface option is the only required option. It is the name of the Kernel
interface to attach to.
--vdev net_af_xdp,iface=ens786f1
The socket will by default be created on Rx queue 0. To ensure traffic lands on this queue, one can use flow steering if the network card supports it. Or, a simpler way is to reduce the number of configured queues for the device to just a single queue which will ensure that all traffic will land on that queue (queue 1) and thus reach the socket. This can be configured using ethtool:
ethtool -L ens786f1 combined 1
5.2.2. start_queue
To create a socket on another queue, first configure the netdev with multiple queues, for example 2, like so:
ethtool -L ens786f1 combined 2
Then, create the socket on one of those queues, for example queue 1:
--vdev net_af_xdp,iface=ens786f1,start_queue=1
5.2.3. queue_count
To create a PMD with sockets on multiple queues, use the queue_count arg. The following example creates sockets on queues 0 and 1:
--vdev net_af_xdp,iface=ens786f1,queue_count=2
5.2.5. xdp_prog
The xdp_prog argument allows for the user to provide a path to a custom XDP program which should be used in place of the default libbpf/libxdp program which simply redirects packets to the sockets. For example:
--vdev net_af_xdp,iface=ens786f1,xdp_prog=/path/to/prog.o
5.2.6. busy_budget
The busy polling feature aims to improve single-core performance for AF_XDP sockets under heavy load. It is enabled by default if the detected kernel version is sufficient ie. >= v5.11. The busy_budget arg sets the busy-polling NAPI budget which is the number of packets the kernel will attempt to process in the netdev’s NAPI context. It can be configured like so:
--vdev net_af_xdp,iface=ens786f1,busy_budget=32
To disable busy polling, simply set the busy_budget to 0:
--vdev net_af_xdp,iface=ens786f1,busy_budget=0
It is also strongly recommended to set the following for optimal performance when using the busy polling feature:
echo 2 | sudo tee /sys/class/net/ens786f1/napi_defer_hard_irqs
echo 200000 | sudo tee /sys/class/net/ens786f1/gro_flush_timeout
The above defers interrupts for interface ens786f1 and instead schedules its NAPI context from a watchdog timer instead of from softirqs. More information on this feature can be found at [1].
5.2.7. force_copy
The force_copy argument allows the user to force the socket to use copy mode instead of zero copy mode (if available).
--vdev net_af_xdp,iface=ens786f1,force_copy=1
5.2.8. use_cni
The EAL vdev argument use_cni is used to indicate that the user wishes to
enable the AF_XDP Device Plugin for Kubernetes with a DPDK application/pod.
--vdev=net_af_xdp0,use_cni=1
Note
When using use_cni, both parameters xdp_prog and busy_budget are disabled as both of these will be handled by the AF_XDP plugin. Since the DPDK application is running in limited privileges so enabling and disabling of the promiscuous mode through the DPDK application is also not supported.
5.2.9. dp_path
The EAL vdev argument dp_path is used alongside the use_cni argument
to explicitly tell the AF_XDP PMD where to find the UDS
to interact with the AF_XDP Device Plugin for Kubernetes.
If this argument is not passed alongside the use_cni argument
then the AF_XDP PMD configures it internally.
--vdev=net_af_xdp0,use_cni=1,dp_path="/tmp/afxdp_dp/<<interface name>>/afxdp.sock"
5.3. Limitations
MTU
The MTU of the AF_XDP PMD is limited due to the XDP requirement of one packet per page. In the PMD we report the maximum MTU for zero copy to be equal to the page size less the frame overhead introduced by AF_XDP (XDP HR = 256) and DPDK (frame headroom = 320). With a 4K page size this works out at 3520. However in practice this value may be even smaller, due to differences between the supported RX buffer sizes of the underlying kernel netdev driver.
For example, the largest RX buffer size supported by the underlying kernel driver which is less than the page size (4096B) may be 3072B. In this case, the maximum MTU value will be at most 3072, but likely even smaller than this, once relevant headers are accounted for eg. Ethernet and VLAN.
To determine the actual maximum MTU value of the interface you are using with the AF_XDP PMD, consult the documentation for the kernel driver.
Note: The AF_XDP PMD will fail to initialise if an MTU which violates the driver’s conditions as above is set prior to launching the application.
Shared UMEM
The sharing of UMEM is only supported for AF_XDP sockets with unique contexts. The context refers to the netdev,qid tuple.
The following combination will fail:
--vdev net_af_xdp0,iface=ens786f1,shared_umem=1 \ --vdev net_af_xdp1,iface=ens786f1,shared_umem=1 \
Either of the following however is permitted since either the netdev or qid differs between the two vdevs:
--vdev net_af_xdp0,iface=ens786f1,shared_umem=1 \ --vdev net_af_xdp1,iface=ens786f1,start_queue=1,shared_umem=1 \
--vdev net_af_xdp0,iface=ens786f1,shared_umem=1 \ --vdev net_af_xdp1,iface=ens786f2,shared_umem=1 \
Secondary Processes
Rx and Tx are not supported for secondary processes due to memory mapping of the AF_XDP rings being assigned by the kernel in the primary process only. However other operations including statistics retrieval are permitted. The maximum number of queues permitted for PMDs operating in this model is 8 as this is the maximum number of fds that can be sent through the IPC APIs as defined by RTE_MP_MAX_FD_NUM.
libxdp
When using the default program (ie. when the vdev arg ‘xdp_prog’ is not used), the following logs will appear when an application is launched:
libbpf: elf: skipping unrecognized data section(7) .xdp_run_config libbpf: elf: skipping unrecognized data section(8) xdp_metadata
These logs are not errors and can be ignored.